Regularization Techniques: A Comparative Analysis of Ridge, Lasso, and Elastic Net Approaches in Predicting Mental Health Consequences Using Mental Health Survey Dataset

Abstract

This study investigated the demographic distribution, predictor relationships, and model performance concerning factors influencing work interference among employees. A Chi-square goodness-of-fit test revealed a significant gender imbalance in the sample, with 60% males and 40% females (χ² = 6.00, p = 0.014), indicating a deviation from an expected equal distribution. Despite this, gender differences had minimal effect on the main outcomes. Variance Inflation Factor (VIF) analysis confirmed the absence of multicollinearity among predictors, with the maximum VIF recorded at 2.10 and the mean VIF at 1.45. Cross-validation of Ridge, LASSO, and Elastic Net regression models produced low Root Mean Squared Error (RMSE) values, with Ridge Regression achieving the best fit (RMSE = 4.74). Pseudo R-squared values ranged between 0.42 and 0.44, highlighting the models' moderate explanatory power. Standardized coefficients identified Job Stress as the most influential predictor, followed by Workload, Support from Supervisor, Work-Life Balance, Organizational Commitment, and Job Autonomy. The findings underscore the critical role of reducing stress and workload to minimize work interference and improve organizational productivity. Recommendations include strategic interventions targeting stress management and balanced work demands, alongside improving supervisory support structures.